Comparing Bivariate Relationships - Math 1

1. An economist analyzes the relationship between two economic indicators using different time periods. Period 1 (2010-2015): $r = 0.78$, $n = 72$ monthly observations. Period 2 (2016-2021): $r = 0.82$, $n = 72$ monthly observations. When comparing these relationships to determine which period shows a more reliable economic relationship, what additional information would be most critical?
   Answer: The economic conditions and external factors present in each period that might affect the stability of the relationship. For determining which period shows a more reliable economic relationship, understanding the economic context and external factors is most critical. Economic relationships can be influenced by market conditions, policy changes, or external shocks. A correlation that holds during stable conditions might be more reliable than one during volatile periods, regardless of the numerical value. This contextual information helps assess the generalizability and stability of the relationship.
2. A biologist compares two models for population growth: Model X yields $r^2 = 0.94$ but shows systematic patterns in residuals, while Model Y yields $r^2 = 0.87$ with randomly distributed residuals. The biologist chooses Model X because of its higher $r^2$ value. What is the most serious issue with this decision?
   Answer: The systematic residual patterns in Model X suggest model misspecification, which undermines the reliability of the high $r^2$ value. When evaluating statistical models, you need to look beyond just the correlation coefficient ($r^2$) and examine residual patterns, which reveal whether your model captures the underlying relationship correctly. The key issue here is that systematic patterns in residuals indicate model misspecification - meaning the model doesn't properly represent the true relationship in the data. When residuals show patterns (like curves, trends, or clustering), it means your model is consistently missing something important about how the variables relate. This makes the high $r^2 = 0.94$ misleading because it suggests the model explains 94% of the variance, but the systematic residuals reveal this explanation is fundamentally flawed. Model Y's randomly scattered residuals indicate it correctly captures the relationship structure, even though it explains less total variance. Looking at the wrong answers: B incorrectly claims lower $r^2$ indicates better stability - that's not what $r^2$ measures. C focuses on statistical significance testing between $r^2$ values, but the real problem is model validity, not whether the difference is statistically significant. D mentions overfitting, but systematic residuals specifically indicate underfitting or wrong model form, not overfitting (which would show up differently in validation testing). Study tip: Always examine residual plots when comparing models. Random, scattered residuals are more important than a high $r^2$ because they confirm your model structure is correct. A lower $r^2$ with proper residuals beats a higher $r^2$ with systematic patterns every time.
3. A data analyst fits both exponential and power models to a dataset and obtains the following results: Exponential model: $\ln(y) = 0.23x + 2.1$ with $r^2 = 0.89$; Power model: $\ln(y) = 1.4\ln(x) + 0.8$ with $r^2 = 0.76$. When determining which transformation is more appropriate, what should be the primary consideration?
   Answer: The choice should depend on examining residual plots and considering the theoretical relationship expected between the variables. While the exponential model has a higher $r^2$ value, choosing between models should not rely solely on this statistic. The most appropriate approach is to examine residual plots to check model assumptions, consider the theoretical relationship between variables, and evaluate how well each model captures the underlying pattern. The difference in $r^2$ values alone doesn't tell us which model better represents the true relationship.
4. Two different laboratories measure the same chemical reaction under identical conditions. Lab A reports: $y = 2.8x + 0.5$, $r^2 = 0.92$, $n = 30$. Lab B reports: $y = 2.9x + 0.3$, $r^2 = 0.88$, $n = 45$. Considering both statistical and practical significance, which laboratory's results should be considered more reliable?
   Answer: Lab B because the larger sample size provides more statistical power, and the regression parameters are very similar to Lab A's results. When evaluating competing regression analyses, you need to balance model fit quality with sample size considerations, while also examining practical significance through parameter similarity. Lab A shows $r^2 = 0.92$ versus Lab B's $r^2 = 0.88$, meaning Lab A's model explains 92% of the variance compared to 88% for Lab B. However, Lab B has a substantially larger sample ($n = 45$ vs $n = 30$), providing greater statistical power. Crucially, the regression parameters are remarkably similar: slopes of 2.8 vs 2.9 and intercepts of 0.5 vs 0.3. This consistency across different labs suggests both are capturing the same underlying relationship, making the practical significance very strong. Choice A is correct because the larger sample size in Lab B provides more reliable parameter estimates and greater generalizability, while the similar regression parameters indicate both labs are measuring the same phenomenon consistently. Choice B incorrectly prioritizes $r^2$ over sample size. While Lab A's fit is slightly better, the difference is modest and doesn't outweigh the statistical advantages of Lab B's larger sample. Choice C makes a flawed assumption that higher $r^2$ in smaller samples indicates stronger relationships. Actually, smaller samples are more susceptible to outliers that can artificially inflate $r^2$ values. Choice D overstates the case by claiming larger samples "always" produce more reliable estimates "regardless" of other factors, which ignores model fit entirely. Remember: When comparing studies, look for the combination of adequate sample size, reasonable model fit, and consistent parameter estimates across different conditions.
5. Two datasets show the following characteristics: Dataset X has a linear regression equation $y = 3.2x + 1.8$ with $r^2 = 0.64$, while Dataset Y has a quadratic regression equation $y = 0.8x^2 - 2.1x + 5.4$ with $R^2 = 0.71$. A student concludes that the quadratic model is better because it has a higher coefficient of determination. What is the most significant flaw in this reasoning?
   Answer: The student didn't account for model complexity; the modest improvement in fit may not justify the additional complexity of the quadratic model. The most significant flaw is that the student didn't consider whether the modest improvement in $R^2$ (from 0.64 to 0.71) justifies the increased complexity of a quadratic model. Good model selection involves balancing goodness of fit with parsimony. A 7 percentage point improvement might not be substantial enough to warrant the more complex model, especially without considering other factors like residual analysis or the principle of Occam's razor.
6. Two researchers study the same phenomenon using different sample sizes and obtain these results: Study 1 (n=25): $y = 4.8x - 12.3$, $r = 0.71$, standard error = 2.8; Study 2 (n=100): $y = 4.2x - 9.7$, $r = 0.68$, standard error = 1.9. When comparing the reliability of these linear models, which factor most strongly favors one study over the other?
   Answer: Study 2 is more reliable because the lower standard error indicates more precise parameter estimates despite the slightly lower correlation. The lower standard error in Study 2 (1.9 vs 2.8) most strongly favors its reliability. Standard error measures the precision of parameter estimates, and Study 2's lower value indicates more precise estimates of the slope and intercept. While Study 2's larger sample size also contributes to reliability, the standard error directly quantifies the precision advantage, making it the most compelling factor.

Start this quiz